Intel B660 Motherboards May Lack PCIe 5.0 Support

[chrcoluk]

PCIe 5 doesn't matter don't worry about it, some board manufacturers are advertising it as needed to fully handle newer gen GPU's, very unlikely that will be the case. The only use case I see for it is super fast storage controllers, so maybe a home lab or DIY NAS setup but that's it.

The GPU PCIe5 slot on Z690 only has one benefit, which is marketing.

 

[Valantar]

I think what they did was smart (if not in an obvious way). If they connected the chipset over PCIe5, the chipset would have had to implement PCIe5, too. Which likely would have resulted in a more power hungry chipset, that would have needed active cooling. With only 16 lanes available, I think it was wise to use them for the graphics slot, because that's what will use PCIe5. Sure, it will provide no benefits at all before Alder Lake is obsoleted, but that the game every time a new PCIe revision makes its way into consumer products.

Also, keep in mind that, unlike AMD, B is not the mid-range chipset, it's the low-end/business oriented. Intel's mid-range is H chipsets.

That might well be true, though I wonder whether four lanes of 5.0 for the chipset would be more power hungry than the current eight lanes of 4.0. IMO, the complexity of 5.0 for the chipset is more of a reason, with the need for signal retimers etc.

There's still the question of whether PCIe 5.0 for GPUs will have any utility in the usable lifetime of these CPUs and motherboards though (by which I mean 5-7 years). Given that current top-end GPUs lose 1% or less from 4.0 x16 to 3.0 x16, it's quite unlikely for another doubling in bandwidth to be useful for quite a long time. After all, that tells us that GPUs aren't really using PCIe bandwidth beyond the speed of 3.0 today, except in a few tiny edge cases. And "beyond 3.0" might mean 1% above its capabilities, or 10% or 20%, and most likely not 100% more like 4.0 already provides - so there's plenty of room to grow still with 4.0.

 

[chrcoluk]

That might well be true, though I wonder whether four lanes of 5.0 for the chipset would be more power hungry than the current eight lanes of 4.0. IMO, the complexity of 5.0 for the chipset is more of a reason, with the need for signal retimers etc.

There's still the question of whether PCIe 5.0 for GPUs will have any utility in the usable lifetime of these CPUs and motherboards though (by which I mean 5-7 years). Given that current top-end GPUs lose 1% or less from 4.0 x16 to 3.0 x16, it's quite unlikely for another doubling in bandwidth to be useful for quite a long time. After all, that tells us that GPUs aren't really using PCIe bandwidth beyond the speed of 3.0 today, except in a few tiny edge cases. And "beyond 3.0" might mean 1% above its capabilities, or 10% or 20%, and most likely not 100% more like 4.0 already provides - so there's plenty of room to grow still with 4.0.

I plan to test my 3080 on a gen 3x4 slot, I be surprised if there is anything more than a few % loss which would be down to less optimal latency on the traces. On gen 3x16 I havent seen it use anything higher than 6% of bandwidth.

 

[Wirko]

The only use case I see for it is super fast storage controllers, so maybe a home lab or DIY NAS setup but that's it.

Whoever does that will also buy a PCIe expansion card and have four M.2 slots available. Sure these cards won't come cheap but other components won't, either.

 

[Bomby569]

There won't be SSDs or GPUs in 3-5 years where this represents a meaningful disadvantage either. If a GPU in 5 years is more than a couple of % slower on PCIe 4.0 than 5.0, then your 2021-era CPU is more likely to hold it back than that PCIe link. And as for storage, until we start seeing applications make better use of the speed we already have available, the only benefits from 5.0 SSDs will be higher benchmark scores.

it's hard to predict the future, but there is direct storage

 

[Wirko]

FYI, none of the m.2 on Z690 boards are pcie 5 anyway. Only the 16x slot is gen5.
I find it rather odd, if anything the gen 5 support is best used on the DMI link, but they didn't do so.

Hey, some Asus boards have 5.0 wires running to M.2 (including this one, bwahaha).

 

[chrcoluk]

it's hard to predict the future, but there is direct storage

True, and I considered that, the problem is the chipset doesnt have gen 5 m.2 slots.

So if a game is on e.g. a gen 4x4 m.2, then you add that much bandwidth in/out to the GPU, it still doesnt need gen 5x16, gen 4x8 would probably be enough.

 

[Wirko]

Intel B660 boards may lack DDR5 support altogether, too. Would anyone be surprised? Both PCIe 5 and DDR5 require better quality PCB materials and possibly more layers. If that brings manufacturing cost up by a handful of dollars, it's only logical for both to be cut off from $80-$150 boards.

 

[renz496]

It is surprising that Intel decided to release PCIe 5 support many years before the consumers might actually see this technology in mass quantities.
Until recently, Intel had no PCIe 4 support, and now it hurries for marketing "advantages", other than any meaningful use of the supported features.

The software is years and years behind, maybe after 5 years, the consumers will see any meaning in going to PCIe 5.

even PCI-E 4.0 are not really needed. at least for regular consumer PC. just how long did we end up being stuck with PCI-E 3? i think if not because of solid state storage that start taking advantage of PCI-E speed and bandwidth we probably did not really move beyond PCI-E 3. but faster and better PCI-E spec also give hardware maker some idea like what AMD did to their mid range GPU: limiting it to x8 only. so don't be surprise when PCIE 5 actually arrive AMD will limiting mid range GPU to x4 only while their flagship at 8x only to maximize profit margin haha.

 

[chrcoluk]

Intel B660 boards may lack DDR5 support altogether, too. Would anyone be surprised? Both PCIe 5 and DDR5 require better quality PCB materials and possibly more layers. If that brings manufacturing cost up by a handful of dollars, it's only logical for both to be cut off from $80-$150 boards.

B660 may well be the way to go? Z690 spending cash on PCB layers that have no practical purpose.

Be interesting to see if there is any high quality B660 boards with reasonable VRMs and 8 SATA ports. If I remember right intel's midrange chipset now allows memory overclocking?

 

[AnarchoPrimitiv]

Hey, some Asus boards have 5.0 wires running to M.2 (including this one, bwahaha).

Just out of curiosity, why wouldn't you link to the motherboard's product page on the Asus website? It seems weird to link to a retailer's site, and in German no less... Not that I think everyone speaks English, but so far, every comment in this thread has been in English so it'd make more sense

 

[Wirko]

Just out of curiosity, why wouldn't you link to the motherboard's product page on the Asus website? It seems weird to link to a retailer's site, and in German no less... Not that I think everyone speaks English, but so far, every comment in this thread has been in English so it'd make more sense

I wanted to point out the price more than anything else, it's the most costliest board I've seen so far.

 

[yeeeeman]

that's ok and good since it will lower prices.

 

[Bomby569]

that's ok and good since it will lower prices.

Dream on, most likeky the gen5 will just sell at a premium

 

[Valantar]

I plan to test my 3080 on a gen 3x4 slot, I be surprised if there is anything more than a few % loss which would be down to less optimal latency on the traces. On gen 3x16 I havent seen it use anything higher than 6% of bandwidth.

Did you see the review I linked in the post you quoted? TPU has tested this already. 3.0x4 is equivalent to 1.0x16, so you'll lose some performance, but it's by no means horrendous. If you're connecting it to the chipset I would expect a bit more loss, but still not a ton.

it's hard to predict the future, but there is direct storage

Sure, but DS will at least at first reduce bandwidth requirements as it allows for transferring compressed rather than decompressed data between the SSD and GPU. It might of course lead to game developers leaning more on just-in-time asset streaming than they do today, but if that happens it's going to take a long, long time to come true. People learning new ways of developing games isn't something that turns around quickly.

Dream on, most likeky the gen5 will just sell at a premium

It absolutely will, just look at Z690 prices, but at least this will help maintain B660 boards at about the same price level of current midrange boards.

 

[Chrispy_]

Pretty sure there's no consumer/client products launched or due to launch within the retail lifespan of B660 that need PCIe 5.0.

Given how expensive PCIe 5.0 makes boards, there's zero reason to add it. Hell, Intel could leave these boards on PCIe 3.0 and still probably get away with it because consumers would likley be happy with cheaper boards, provided that was a tradeoff that actually trickled down to them. Want more PCIe bandwidth? Buy a higher-end chipset and motherboard....

 

[Bomby569]

Pretty sure there's no consumer/client products launched or due to launch within the retail lifespan of B660 that need PCIe 5.0.

Given how expensive PCIe 5.0 makes boards, there's zero reason to add it. Hell, Intel could leave these boards on PCIe 3.0 and still probably get away with it because consumers would likley be happy with cheaper boards, provided that was a tradeoff that actually trickled down to them. Want more PCIe bandwidth? Buy a higher-end chipset and motherboard....

what's the cost difference in making a motherboard?

 

[bug]

...

Sure, but DS will at least at first reduce bandwidth requirements as it allows for transferring compressed rather than decompressed data between the SSD and GPU. It might of course lead to game developers leaning more on just-in-time asset streaming than they do today, but if that happens it's going to take a long, long time to come true. People learning new ways of developing games isn't something that turns around quickly.

...

On the other hand, developers with resources to waste are like dogs with bones

 

[Valantar]

On the other hand, developers with resources to waste are like dogs with bones

Oh, absolutely. It'll just be 3-5 years until those developers can put out a game that's built on this infrastructure to begin with, and even then I think they'll struggle to exceed what a good PCIe 3.0 drive can deliver.

 

[Bomby569]

Oh, absolutely. It'll just be 3-5 years until those developers can put out a game that's built on this infrastructure to begin with, and even then I think they'll struggle to exceed what a good PCIe 3.0 drive can deliver.

It's a very specific use, you can't design a game to take advantage of faster drives and make a setting to turn it off. And so many people still used HDD's that it can't happen like that. It has to be done at the OS level, i think.

 

[bug]

Oh, absolutely. It'll just be 3-5 years until those developers can put out a game that's built on this infrastructure to begin with, and even then I think they'll struggle to exceed what a good PCIe 3.0 drive can deliver.

You're probably underestimating how easy it is to request a ton of textures every second or so. Just because you can

 

[Keullo-e]

This is not surprising to be honest. The B660 is meant to be a budget oriented chipset, and if removing PCI-E 5.0 support helps lower cost, then it is worth omitting PCI-E 5.0 support. In fact, PCI-E 4.0 is still plenty of bandwidth to go around for GPU. SSDs can benefit more from PCI-E 5.0, but most people won't see the benefit of the higher bandwidth since they are not moving big files frequently. When it comes to loading of application/ games, I still don't see tangible benefit moving from a fast PCI-E 3.0 to 4.0 SSD.

No doubt as with RTX 3080 the difference between 3.0 x16 and 4.0 x16 is about 1%. Even 3.0 x16 is more than enough for GPUs.

NVIDIA GeForce RTX 3080 PCI-Express Scaling

NVIDIA Ampere finally brings PCI-Express 4.0 support to the high-end graphics market. The new interface promises twice the bandwidth of PCI-Express 3.0. We've setup an AMD Ryzen 3900XT system to test how various PCIe generations and lane widths affect gaming performance.

www.techpowerup.com

 

[Valantar]

It's a very specific use, you can't design a game to take advantage of faster drives and make a setting to turn it off. And so many people still used HDD's that it can't happen like that. It has to be done at the OS level, i think.

OS changes won't affect anything in how games load files - they can create new ways of doing so, but can't enforce them on programs not made to do so (unless you want to run the risk of tons of instability). DirectStorage is precisely such an attempt at creating a new way of doing this, by standardizing GPU-based on-the-fly decryption and direct SSD-to-VRAM data transfers. Still, it's only a framework. And the structure of the files being loads matters a lot - the classic example from the PS5 launch was how legacy games made for HDDs expect high seek times and thus massively duplicate assets to avoid the drive having to seek for them every time they are needed. This is obviously not necessary if the game is developed with an SSD as the baseline - but doing so will also potentially harm performance for those failing to meet the base spec. Hence why these developments are so slow in trickling into actual games. Consoles have a lot of defining power here, as they essentially set a baseline per generation that all PC ports maintain unless the port developer wants to drastically increase their workload. So now that we have SSD-only consoles, we will start to see games developed expecting SSDs even on PCs. But due to how data is typically accessed in actual software, even a SATA SSD will likely be sufficient for this.

 

[Punkenjoy]

Right now, the main driver for higher PCI-E speed are SSD on servers. SSD can saturate those bandwidth.

For GPU, right now, Devs make their games so they can run well on current GPU and motherboard. Each PCI-E revision double the bandwidth. Let say a game is made that it require PCI-E gen 4, well that would be disastrous for previous Gen and they will cut out a huge amount of player.

That is also another reason why Direct Storage and Sampler Feedback might not come to game engine anytime soon. Once they will be there on the market, current hardware might be outdated. And anyway, it will never be a problem to move data from a 4x port to a 16x port. there shouldn't be a bottleneck anyway.

Higher PCI-E bandwidth might at some point lead to graphic cards not requiring a full 16x slot. a 4x PCI-E Gen 5 slot is as fast as a PCI-E Gen 3 16x slot. A little bit like the lower end AMD cards that only support 8x. But also, it's not that useful to have 64 GB/s of bandwidth on a PCI-E gen 5 16x slot if you are barely able to fill it with current memory. DDR5 help there, but still.

smaller bus mean less trace witch mean less cost.

But right now, PCI-E 4 and 5 are mostly just useful on large server and on workstation for people that have require a lot of I/O power. Not having Gen5 on those board is not a big deal at all.

 

[Valantar]

That is also another reason why Direct Storage and Sampler Feedback might not come to game engine anytime soon. Once they will be there on the market, current hardware might be outdated. And anyway, it will never be a problem to move data from a 4x port to a 16x port. there shouldn't be a bottleneck anyway.

That's not what DS does. DS allows for decompression on the GPU, meaning the data path becomes SSD -> VRAM instead of SSD -> RAM -> CPU (decompression -> VRAM. Of course you could always store uncompressed game assets on the SSD but ... yeah, we have enough 100+GB games, thank you. And there's no reason to expect future hardware to not be compatible with DS - it's pretty hardware agnostic beyond the GPU needing to be able to decompress specific compression types. This isn't going away any time soon.

I completely agree that lane count reductions are a great potential benefit, but sadly I don't think we're going to see that any time soon. I would love to see affordable PCIe 4.0x2 SSDs come to market, and as you say, half the lanes for a GPU that doesn't need more is only a good thing. But both of those things risks hurting backwards compatibility (limiting your SSD to 3.0x2 or your GPU to 3.0x8 in the wrong situation), which IMO is why we'll never see them come to market - there is too large of an install base for older I/O standards, and nobody wants to exclude that market. Nor do motherboard makers want to do the same by limiting the value of hardware you bring with you from a previous build (if the m.2 slots or PEG slot only had half the lane, suddenly you lose performance through upgrading if you're keeping an SSD or GPU).